1

I was talking with a colleague who told me that at the time of making logistic regression across a population did not have to worry about assumptions such as multicollinearity, because when analyzing the entire population estimated betas are "actual parameters" and since to "multicollinearity only affects the significance of the estimates" not matter in this case.

This got me thinking, could it be true? If so, under what circumstances?

I was reading some articles that talk about when multicollinearity is not important, but does not mention anything about population analysis.

On the other hand I was reading some answers on population analysis and understand that suggest treating this type of regressions as if a large sample from a super-population analyzed.

What would be the best position on this type of analysis (population) and multicollinearity (excluding the cases mentioned in the articles linked)?

Community
  • 1
Galled
  • 145
  • 6
  • Of course also in a "population model", effects from highly correlated variables cannot be easily interpreted in an isolated way. – Michael M Aug 05 '14 at 14:21
  • @MichaelMayer But, what you say about the validity of the model? To make predictions, to make decisions based on it? – Galled Aug 05 '14 at 14:24
  • I have never discovered an equivalent to the variance inflation factor for binomial outcome regression models. Do you have ideas about this? – Alexis Aug 05 '14 at 18:51

0 Answers0